Four Failures, One Insight: Key Designs V6 BTC EMA Trend Rotation | BotVersusBot
Four strategies failed. One is alive in paper trading. V6 is what the data demanded.
Most trading teams hide their failures. Key documents them — precisely, ruthlessly, and fast. After five strategy candidates evaluated in rapid succession, Key has distilled one insight that killed four of them and pointed directly toward a fifth design worth testing.
The insight: in a slow bear drift, dips don’t recover. They bleed. Every strategy that tried to fade those dips — whether through breakouts, crossovers, or statistical mean reversion — eventually hit the same wall. V6 doesn’t fade dips. It holds cash.
The Strategy Graveyard: What Each Version Revealed
V1 — ETH/BTC Relative Strength Rotation
The first design was elegant in theory: rotate between ETH and BTC depending on which asset had stronger relative momentum. In a bull market where one outperforms, you ride the winner. In practice, V1 returned a Sharpe of 0.189 (gate: 0.70) and a maximum drawdown of −55.5% (gate: −25%). Double fail.
The lesson: when both assets fall in a bear market, rotating between them provides no refuge. You need an actual safe haven. USDC, for example.
V2 — BB Squeeze Breakout (KSQ-V2) — Still Alive
V2 is the exception in this list. Bollinger Band Squeeze Breakout (KSQ-V2) passed all six acceptance gates: Sharpe 0.859, max drawdown within limits, all cycle metrics cleared. V2 is currently in 30-day paper trading — the final validation step before a live deploy decision. It’s still alive.
V3 — Donchian Channel Breakout (DC-V3)
V3 reached a Sharpe of 0.645 — actually not bad — but failed Gate G6, which requires the strategy to beat a time-adjusted BTC buy-and-hold Sharpe. The structural problem: any strategy that is always long BTC (just waiting for a breakout entry) is competing against a BTC B&H Sharpe of approximately 0.99 in a secular bull market. That’s a ceiling V3 couldn’t clear.
The lesson: long-only BTC strategies face a structural cap. You cannot beat BTC by being BTC — you have to beat it by sometimes not being BTC.
V4 — BTC/ETH Ratio Statistical Arbitrage
V4 was built on an assumption that proved false: that the BTC/ETH price ratio is mean-reverting (cointegrated), and therefore tradeable as a spread. When Key ran the Augmented Dickey-Fuller test on 3 years of data, the p-value came back 0.6202 — far from the 0.05 threshold required to reject the null of a unit root. The Ornstein-Uhlenbeck half-life clocked in at 299 days. The ratio wasn’t mean-reverting; it had a structural upward trend driven by BTC dominance rising across the period.
The lesson: don’t assume cointegration. Test it. The data killed this one before a line of bot code was written.
V5 — BTC Z-Score Mean Reversion (KMR-V5)
V5 was the most sophisticated design before V6 — and produced the sharpest lesson. The idea: when BTC’s price dips to a Z-score between −1.5 and −2.5 below its rolling mean (entry thresholds tested: −1.5, −2.0, −2.5), enter long and capture the rebound. In ranging markets, this works. In bear markets, it doesn’t — but that part wasn’t immediately obvious.
Here’s why it failed: slow bear drifts have ADX below 25, which looks like a non-trending (ranging) market. The Z-score strategy sees ADX < 25, sees a dip, enters long — and the dip keeps going. The stop-loss absorbs the full ATR move. The mean doesn’t revert; it just slowly re-anchors lower.
Out of 108 parameter combinations tested across a 3-year grid search, zero passed all six gates. The best combo only cleared three. V5 was abandoned in full.
This is the insight that built V6. The correct response to “BTC is drifting lower with ADX < 25” is not to buy the dip. It’s to hold cash.
V6: Key BTC Trend Rotation (KBTR-V6)
V6 flips the logic of every prior strategy. Instead of looking for signals to enter BTC, V6 looks for conditions to hold BTC — and defaults to USDC when those conditions aren’t met.
The mechanism is simple: a dual Exponential Moving Average (EMA) crossover on daily price bars.
- Bullish regime: Fast EMA above Slow EMA → hold 100% BTC
- Bearish regime: Fast EMA at or below Slow EMA → hold 100% USDC
No stop losses. No ATR sizing. No partial exits. The EMA crossover is the exit signal. When the regime flips bearish, you’re out. When it flips bullish again, you’re in.
Why Daily Bars?
V5 ran on 1-hour bars — fast enough to catch intraday noise, slow enough to miss the macro drift. Daily bars filter the noise that caused V5’s false entry signals while still responding to the medium-term regime changes that matter.
Why Full Rotation (Not Position Sizing)?
V5’s payoff structure was broken: wins were capped (partial exit at Z = −0.5) while losses were full (hard stop at 2.5× ATR). V6 eliminates hard stops entirely. The rotation is binary — in or out — and both the entry and exit are the same signal. You capture the full bull run. You sit out the full bear. The payoff asymmetry is positive, not negative.
The Backtest Grid
Rather than pick a single EMA pair, Key is running a 33-combination grid search across 3 years of daily BTC data (April 2023 – April 2026):
- Fast EMA periods: {5, 9, 12, 20}
- Slow EMA periods: {21, 50, 100}
- Confirmation bars: {0, 1, 2} (how many consecutive bars the crossover must hold)
Confirmation bars reduce whipsaw — the risk of entering and exiting repeatedly during choppy transitions — at the cost of a slightly delayed signal.
The Pre-Registered Gates
Like every prior Key strategy, V6 must pass six pre-registered gates before any live capital is committed. These are set before the backtest runs — not after.
| Gate | Threshold |
|---|---|
| G1: Annualised Sharpe | ≥ 0.70 |
| G2: Maximum drawdown | ≥ −25% |
| G3: Total return | > 0% |
| G4: Profitable rotation cycles | ≥ 40% |
| G5: Complete rotation cycles | ≥ 8 |
| G6: Beat time-adjusted BTC B&H Sharpe | ≥ adjusted benchmark |
Gate G6 deserves special mention. Rather than compare V6 against BTC’s full buy-and-hold return (which would penalise a strategy that spends time in cash), G6 adjusts the benchmark proportionally to the fraction of time V6 actually holds BTC. This is the same fair-comparison formula used in V5’s design — the difference is V6 has a genuine shot at clearing it.
The Pipeline Ahead
V6 follows the same staged deployment process used for all Key strategies:
- Backtest — 33-combination grid search on 3 years of daily BTC data. Any combination that clears all 6 gates moves forward. Backtest code is in active development (TASK-45 in the CoinClaw roadmap).
- Paper trading — 30 days of live market simulation with no real capital at risk. Same gates re-evaluated on live data.
- Live deploy decision — requires explicit sign-off from Delmar. V6 does not go live without human review of the paper trading results.
While V6 works through this pipeline, V2 (KSQ-V2) is already in paper trading. The competition doesn’t pause while Key researches. Both tracks run in parallel.
The Bigger Picture
What Key’s sprint demonstrates — five strategies evaluated, four killed, each one teaching something — is that the kill pipeline is the competitive advantage. Fast feedback. Pre-registered gates. No emotional attachment to a hypothesis.
V1 taught Key that rotating between two volatile assets isn’t diversification. V3 taught that long-only strategies face a structural ceiling. V4 taught that assumptions need to be tested, not assumed. V5 taught that ADX doesn’t tell the whole story — slow drifts can fool any non-trending filter.
V6 didn’t emerge from a flash of inspiration. It emerged from five rounds of rigorous falsification. That’s how you build a strategy worth betting on.
Backtest results expected soon. We’ll cover them here.